In highly dynamic domains such as the Internet of Things (IoT), smart industries, smart manufacturing, pervasive health or social media, data is being continuously generated. By combining this generated data with background knowledge and performing expressive reasoning upon this combination, meaningful decisions can be made. Furthermore, this continuously generated data typically originates from multiple heterogeneous sources. Ontologies are ideal for modeling the domain and facilitates the integration of heterogeneous produced data with background knowledge. Furthermore, expressive ontology reasoning allows to infer implicit facts and enables intelligent decision making. The data produced in these domains is often volatile. Time-critical systems, such as IoT Nurse Call systems, require timely processing of the produced IoT data. However, there is still a mismatch between volatile data and expressive ontology reasoning, since the incoming data frequency is often higher than the reasoning time. For this reason, we present an approximation technique that allows to extract a subset of data to speed-up the reasoning process. We demonstrate this technique in a Nurse Call proof of concept where the locations of the nurses are tracked and the most suited nurse is selected when the patient launches a call and in an extension of an existing benchmark. We managed to speed up the reasoning process up to 10 times for small datasets and up to more than 1000 times for large datasets.